Demand Poll to Coax Along Discovery

When NNM isn’t discovering devices fast enough to suit your needs, you need to intervene. NNM may have backed off its discovery aggressiveness, or the netmon configuration checks may be scheduled for a time later than you care to wait. Usually, you have a good idea which device has the necessary information and can direct NNM to poll the node at once. The “Poll Node” menu item, formerly called “Demand Poll,” brings up a GUI that is equivalent to the command line nmdemandpoll. It forces netmon to immediately interrogate the device for its basic configuration, interface table, routing table, and ARP cache. This information usually allows NNM to discover additional devices and topology. However, there are wrinkles to this approach.

Perhaps netmon is already busy harvesting a large ARP cache on a slow router, in which case it will be pointless to poll it. The netmon SNMP queues may be nearly full, again rendering intervention moot. If local SNMP community strings are wrong, the only helpful intervention is to correct them. The NNM system itself may be temporarily overloaded, robbing netmon and the database daemons of needed CPU cycles.

Polling a node is more potent than pinging it. Recall that netmon listens on a raw ICMP socket for new IP addresses in the management domain, and schedules a poll anyway. A manual poll simply changes when netmon polls the node. Once a device is experiencing a poll, the device CPU utilization will rise dramatically, accompanied by a spike in network traffic.

Network maintenance is often scheduled over a weekend. New switches may be installed and old ones removed. Network traffic is low during the off hours—ARP caches may be drained and autodiscovery may have backed off. It may be prudent to write a UNIX cron job to sweep the network electronics with an nmdemandpoll script.

Note that NNM is very stubborn about removing information from its database. NNM may retain old interfaces for a reconfigured or replaced switch even while newly discovered interfaces are added. Polling the switch won’t help. Deleting the switch and pinging it is often the quickest way to fix this because the switch will be rediscovered in its current configuration.